In this paper we consider the problem of recommending modes of transport to users in an urban setting. In particular, we build on our past work in which a general framework for activity recommendation is proposed. To model the personal preferences and habits of users, the framework uses a sequence-based approach to capture the order as well as the context associated with user activity patterns. Here, we extend this work by introducing a machine learning approach to learn and take into account the natural variations in the regularity and repetition of individual user behaviour that occur. We demonstrate the versatility of our recommendation framework by applying it to the transport domain, and an evaluation using a real-world (mode of transport) dataset demonstrates the efficacy of the approach.